Breaking

CoreWeave Deploys NVIDIA Vera Rubin NVL72 at Production Scale; Cognition Reports 4.8x Token Throughput Gain

October 4, 2026 at 19:01 EDT

Foundation Models
Infra & Chips
AI Agents

CoreWeave has begun operating NVIDIA's Vera Rubin NVL72 GPU cluster at production scale, with AI coding startup Cognition as its first customer.

INFRASTRUCTURE MILESTONE

Vera Rubin Enters Production: 4.8x Faster Inference

CoreWeave deploys next-gen NVIDIA GPU with Cognition as first production customer. Massive gains in token throughput and energy efficiency across inference and reasoning workloads.

4.8x

Token Throughput (SWE-2)

10x

Energy Efficiency / MW

3.8x

RL Workloads

Performance Gain Visualization: Vera Rubin vs. Blackwell

Blackwell

1x baseline

Vera Rubin

4.8x gain

Token throughput for SWE-2 inference workloads in real production benchmarks conducted by Cognition on CoreWeave infrastructure, September 2026.

✓ COGNITION'S ADVANTAGE

48% faster token generation
More concurrent agent sessions per GPU
Lower cost per inference request

→ KEY CONSIDERATION

Vera Rubin still in early rollout
Limited customer availability
Represents long-term shift to inference

Continue reading

The rest of this article is for AI News Blitz readers. Choose an option below to keep reading.